Augmented patient-specific functional medical imaging by implicit manifold learning.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Machine Learning

Background:

• Magnetic resonance angiography (MRA) and magnetic resonance imaging (MRI) are crucial for vascular assessment.
• Accurate noninvasive blood pressure measurement in cerebral arteries remains a challenge.
• Current methods often involve costly and complex fluid dynamics models.

Purpose of the Study:

• To develop a machine learning-based method for noninvasive prediction of blood pressure in cerebral arteries.
• To enrich MRA and MRI data using convolutional neural networks.
• To address limitations in patient-specific data availability for predictive modeling.

Main Methods:

• A convolutional neural network was trained on a synthetic arterial network database.
• Implicit manifold learning was employed to handle missing patient-specific input variables.
• The model links arterial geometry and mechanical properties to blood flow dynamics.

Main Results:

• The trained neural network can predict noninvasive blood pressure in cerebral arteries in near real-time.
• Implicit manifold learning successfully adapted the model to available medical measurements.
• Machine learning provides a viable, less expensive alternative to traditional inversion methods.

Conclusions:

• Machine learning, particularly convolutional neural networks with implicit manifold learning, shows significant potential for noninvasive hemodynamic assessment.
• This approach offers a cost-effective and efficient alternative to sophisticated fluid-structure interaction models.
• The method advances the enrichment of MRA and MRI data for clinical applications.